Partitioned moving least-squares modeling of an automatic zoom lens camera

The accuracy of machine vision systems is highly dependent on the correct estimates of the camera intrinsic parameters. This precision is needed in numerous applications like telepresence and robot navigation. In this work, a new technique is proposed, based on the moving least-squares (MLS) approach, to model the intrinsic parameters of an automatic zoom lens camera system. The key issue is to generate the polynomial functions offline using MLS. Then, by employing a surface tessellation algorithm, each parameter is subdivided into several non-overlapping regions each of which will be approximated using a polynomial function. Hence, the intrinsic parameters can now be estimated by simple evaluation of the corresponding functions. Compared to the previous version of the MLS algorithm, the proposed method requires less amount of computations to estimate the parameters without leading to a significant degradation of the results.